private: Avoid unnecessary conversions in shadow computation

The type was changed from double to float for no apparent reason
when moving code in commit 5060081d.

https://bugzilla.gnome.org/show_bug.cgi?id=788627